TDWG2026
Home / Talks / SAL B
talk · Tuesday 22 September · 16:45–⁠16:50 · SAL B

kabr-tools: A Modular Workflow for Automated Video-Based Behavioral Monitoring Across the Autonomy Spectrum

Jenna Kline · Bots, Bits, and Biodiversity

Recording time 8:11:43–8:14:41Open on Vimeo ↗

Summaries and key moments are machine-produced. Check the recording for the speaker’s exact words.

The short versionkabr-tools offers a modular, model-agnostic pipeline from drone or camera-trap video to ecological behaviour analysis.

Overview

What this was about

In a pre-recorded lightning talk, Jenna Kline introduced kabr-tools, which grew out of the KABR (Kenyan Animal Behavior Recognition) drone video dataset collected at Mpala Research Centre in 2023 (about 10 hours of footage, 900+ annotated videos, ~200,000 Hugging Face downloads). The modular workflow runs from drone or camera-trap video collection through model-agnostic detection, tracking and behaviour annotation with manual oversight, to ecological analyses such as time budgets, behaviour transitions and individual behaviour patterns, and has been used for African wild dog studies.

Why it matters. Helps ecologists process video datasets too large for manual analysis while choosing their level of automation.

Key ideas

In the room

  • KABR dataset: three species, about 10 hours of footage, over 900 annotated behaviour videos, about 200,000 downloads on Hugging Face.
  • Workflow stages: detection, tracking and behaviour annotation, all model-agnostic with adjustable autonomy.
  • Supports time budgets, behaviour transitions, land-use comparisons and individual identity tracking.
  • Used by collaborators for camera trap and drone studies of African wild dogs; scripts, data and preprint available online.
Jump in

Notable moments

In their words

Transcript

Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.

Read the transcript ↓
Loading transcript…
Keep exploring

Related talks.

See the map ↗
talk · 22 Sept · 09:45–⁠09:57 · SAL A

Cross-Modal Coordination for Biodiversity Monitoring at Scale: Lessons from SmartWilds

Coordinating multiple sensor modalities with AI at the far edge, informed by synchronised multimodal datasets like SmartWilds, can scale and adapt biodiversity monitoring.

Jenna Kline
talk · 24 Sept · 14:30–⁠14:45 · SAL C

FAIR² Drones: An AI-Ready Standard for Cross-Domain Wildlife Drone Dataset

A shared dataset card that captures ecological, robotic and computer vision metadata makes costly drone datasets reusable across disciplines.

Jenna Kline
talk · 24 Sept · 14:15–⁠14:30 · SAL C

Camtrap DP: extending data standard designed for camera trapping research to support AI workflows and new data sources

Camtrap DP fits data from any stationary sensor, and converging community feedback shows how to extend it for AI provenance and multimodal monitoring.

Karolina Kuczkowska
Programme abstract

kabr-tools: A Modular Workflow for Automated Video-Based Behavioral Monitoring Across the Autonomy Spectrum

Jenna Kline

Quantifying animal behavior from video footage is central to conservation and welfare assessment, but drone and camera-trap deployments now generate datasets that are too large to process manually. Existing tools support individual stages of video-to-behavior processing, yet ecologists lack practical guidance on which stages can be reliably automated and which require manual review. We present kabr-tools, an open-source modular workflow for video-based behavioral monitoring that supports a spectrum of autonomy, from fully manual annotation to fully automated processing. Components for detection, tracking, and classification are interchangeable and taxon-agnostic, producing structured outputs that align with biodiversity data standards. We evaluate the workflow against expert annotation at each pipeline stage and trace how errors propagate into downstream behavioral estimates, using drone footage of Grevy's zebras, Plains zebras, and reticulated giraffes at Mpala Research Centre, Kenya, and drone and camera trap footage of African wild dogs at The Wilds, Ohio. Detection is the primary bottleneck in both systems, since once individuals were reliably detected, tracking and classification added limited additional error for common behaviors. Automated classification achieved 91.3-F1 for the three most frequent ungulate behaviors, declining to 50.3-F1 for rare categories. A camera-source ablation confirmed that tail-class failure reflects insufficient training examples rather than a modality-specific limitation. This error characterization positions animal behavior studies on the autonomy spectrum, given their research question and required precision, and highlights where standardized, machine-actionable behavioral outputs depend on targeted human review.

Submitted to the conference programme before the talk, so the talk itself may differ.